Faster super-resolution imaging for an inspection microscope maker
Super-resolution imaging five times faster at the same optical quality. The autoloader runs twice as fast and one defect analyzer about 200 times faster.
Stack and topics
| Industry | Industrial inspection and life science instruments |
|---|---|
| Region | Europe |
Context
The client and the problem
Our client makes an inspection and control system that combines software and super-resolution microscopes. R&D labs and manufacturers use it for flexible image processing, reproducible analysis and inspection of wafers for defects, in fields from DNA sequencing and genome therapy to aerospace, automotive, pharma and advanced materials.
The client replaced its stop-and-go cameras with much faster and more powerful ones. The scanning procedure and the image processing had to change to keep up, without losing optical quality.
What we built
What did we build?
C++ and machine learning engineers from our team rebuilt the scanning and processing path and added support for new hardware.
- Continuous scanning. The camera now shoots on the move, without pauses, instead of stopping for every frame.
- Multithreaded image processing. A multithreaded pipeline in place of the sequential one, with CUDA on GPU.
- New hardware support. A new light source, focusing system, stage controller and confocal imager, plus the camera support system.
- Improved algorithms. The image positioning algorithm, the autoloader and the defect analyzers built on image processing and ML.
How we worked
Team and process
The team was made of C++ developers and ML engineers working with the client's hardware. Support continued after the main release, so the client could integrate and apply a range of new hardware.
The client agreed to the case being described without its name.
Facts
Project facts
| Fact | Detail |
|---|---|
| Client | A maker of super-resolution microscopes for inspection and life science |
| Region | Europe |
| Our role | Scanning process, image processing pipeline, hardware support, defect analyzers |
| Team | C++ developers and ML engineers |
| Stack | C++, CUDA, GPU, machine learning, image processing |
Result
Where it stands
Super-resolution imaging is five times faster than before, at the same optical quality. The autoloader runs twice as fast and one defect analyzer about 200 times faster.
Next step
Building something similar?
A 30 minute call with Oleg. You leave with a clear view of scope, risks and the next step.
Oleg Pun
CEO, Vareger
